International orthopaedics · 2026
Anticipating femoral shaft nonunion at 1-3 months: a multimodal framework with automated X‑ray segmentation despite metal implants.
СтатьяНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Yang Puxin, Du Xiaofeng, Liu Zixuan, Liu Hao, Li Chengkai, Wu Huizhao, Zhu Yunhao, Ji Yukai, Yang Ziqi, Hou Zhiyong, Xiang Wei, Chen Wei
Аннотация
BACKGROUND: Nonunion is a costly and serious complication in trauma surgery. The RUST score can assess tibial healing on radiographs but offers no prediction of nonunion at nine to 12 months. A prediction model integrating demographics, blood tests, and radiomics fills this gap and brings clinical value.
METHODS: We collected 1-3 month postoperative X-ray, patient data, and blood tests across three centers including 550 patients and 2,344 X-ray. 13 predictive features were selected via two-stage LASSO with group-wise preselection. A U-Net based segmentation framework standardized bone and implant masks under heterogeneous imaging conditions, enabling robust radiomics extraction. ExtraTrees, LightGBM, SVM, CatBoost, Random forest, Gradient boosting, XGBoost, HistGradientBoosting, MLP, Logistic Regression, ElasticNet, Lasso, AdaBoost, Gaussian NB, and Decision Tree were employed to develop NU model.
RESULTS: A total of 492 patients formed the training cohort and 58 patients formed the external validation cohort. An automated segmentation model achieved mean Dice coefficients of 0.947 for bone/callus and 0.921 for implants. LASSO selected 13 predictors: five radiomic features, seven blood tests, and one clinical variable. The ExtraTrees-based NU model yielded an AUC of 0.891±0.021 in the training cohort, with good calibration and net benefit on decision curve analysis. In external validation, AUC was 0.840. The model enables early nonunion risk stratification within three months after surgery.
CONCLUSION: The NU model is the first automated multimodal prediction model integrating radiomics, blood tests, and clinical data to predict femoral shaft nonunion within three months after intramedullary nailing, providing a technically scalable solution for early nonunion risk stratification before conventional imaging becomes diagnostic.
METHODS: We collected 1-3 month postoperative X-ray, patient data, and blood tests across three centers including 550 patients and 2,344 X-ray. 13 predictive features were selected via two-stage LASSO with group-wise preselection. A U-Net based segmentation framework standardized bone and implant masks under heterogeneous imaging conditions, enabling robust radiomics extraction. ExtraTrees, LightGBM, SVM, CatBoost, Random forest, Gradient boosting, XGBoost, HistGradientBoosting, MLP, Logistic Regression, ElasticNet, Lasso, AdaBoost, Gaussian NB, and Decision Tree were employed to develop NU model.
RESULTS: A total of 492 patients formed the training cohort and 58 patients formed the external validation cohort. An automated segmentation model achieved mean Dice coefficients of 0.947 for bone/callus and 0.921 for implants. LASSO selected 13 predictors: five radiomic features, seven blood tests, and one clinical variable. The ExtraTrees-based NU model yielded an AUC of 0.891±0.021 in the training cohort, with good calibration and net benefit on decision curve analysis. In external validation, AUC was 0.840. The model enables early nonunion risk stratification within three months after surgery.
CONCLUSION: The NU model is the first automated multimodal prediction model integrating radiomics, blood tests, and clinical data to predict femoral shaft nonunion within three months after intramedullary nailing, providing a technically scalable solution for early nonunion risk stratification before conventional imaging becomes diagnostic.